Terrestrial and Remote Indexes to Assess Moderate Deﬁcit Irrigation in Early-Maturing Nectarine Trees

: Monitoring plant water status is relevant for the sustainable management of irrigation under water deﬁcit conditions. Two treatments were applied to an early-maturing nectarine orchard: control (well irrigated) and precise deﬁcit irrigation (PDI, based on soil water content thresholds). Moderate water deﬁcits generated by PDI were assessed by comparing terrestrial: stem water potential ( Ψ stem ) and gas exchange parameters, with remote: canopy temperature, normalized di ﬀ erence vegetation (NDVI), and soil adjusted vegetation index (SAVI), plant water status indicators. The Ψ stem was the only indicator that showed signiﬁcant di ﬀ erences between treatments. NDVI and SAVI at the postharvest period were appropriate indexes for estimating winter pruning, although they did not serve well as plant stress indicator. Vapor pressure deﬁcit along with Ψ stem values were able to predict remote sensing data. Ψ stem and canopy to air temperature di ﬀ erence values registered the highest signal intensity and NDVI the highest sensitivity for detecting water deﬁcit situations. The results suggest that care should be taken when using instantaneous remote indicators to evaluate moderate water deﬁcits in deciduous fruit trees; more severe / longer water stress conditions are probably needed. The proposed PDI strategy promoted water saving while maintaining yield, and could be considered a promising tool for semi-arid agrosystems.


Introduction
Efficient irrigation of horticultural crops using technologies that enable the better husbandry of scarce freshwater resources has been increasingly studied in recent years, because irrigated agriculture is the largest user (up to 70%) of freshwater worldwide [1]. This situation might worsen in coming years as the world's population is expected to increase by 30% by 2050 coupled with forecasted climate change and competition with urban, touristic, and industrial activities [2,3].
Among irrigated woody crops, peach and nectarine trees have particularly high irrigation requirements, especially during dry and hot seasons [4,5], when irrigated orchards are frequently subjected to drastic reductions in the water supply. Indeed, this situation is aggravated in early-maturing cultivars with their high water needs during the summer postharvest interval [6,7]. Since irrigation is essential to ensure optimal yield, it is imperative to develop efficient irrigation strategies for peach and nectarine orchards by means of irrigation scheduling based on the requirements of the plants and optimal water productivity.
To optimize water use and save water, deficit irrigation (DI) practices can be used since they minimize any impact on fruit yield and quality, while reducing excessive vegetative growth [2,8]. Average θv values in the monitored soil profile (0-0.5 m) were used to calculate the relative extractable water (REW), defined by the equation proposed by Granier [45]: where R (mm) is the actual soil water content, Rmin (mm) the minimum soil water content measured in dry conditions, and Rmax (mm) the maximum soil water content obtained in each probe. In this study, the values of Rmax and Rmin were normalized with the θv at field capacity and permanent wilting point of 29% and 14%, respectively.

Plant Water Status and Gas Exchange
Plant water status was estimated by measuring stem water potential (Ψstem) at midday (12:00 h solar time) using a pressure chamber (Soil Moisture Equipment Corp. Model 3000). One leaf was selected from each replicate trees from both irrigation treatments (n = 4). Leaves were placed in plastic bags covered with aluminium foil for at least 2 h prior to the measurements, which were carried out every week from April to October following the recommendations of [16,46,47].

Plant Water Status and Gas Exchange
Plant water status was estimated by measuring stem water potential (Ψ stem ) at midday (12:00 h solar time) using a pressure chamber (Soil Moisture Equipment Corp. Model 3000). One leaf was selected from each replicate trees from both irrigation treatments (n = 4). Leaves were placed in plastic bags covered with aluminium foil for at least 2 h prior to the measurements, which were carried out every week from April to October following the recommendations of [16,46,47].

Airborne Campaign
A flight campaign was carried out by Drónica Servicios Aéreos, S.L.L., on 19 July 2017 using a VIS-NIR multispectral and thermal camera installed in an UAV (DJI S900 model). Two flights were conducted at approximately 120 m of altitude over the experimental plot: the first one at around 10:00 GMT (t1) and the second at 12:00 GMT (t2). For this study the autopilot was used, following the waypoints of a flight plan created using flight planner software (Pix4D). The UAV was equipped with a GPS receiver, altimeter, wind meter, and a digital camera that was electronically triggered by the autopilot system to acquire images at the correct positions. The multispectral camera used was a Parrot Sequoia+ (Parrot Co. Ltd., Paris, France), which measures 59 mm × 41 mm × 28 mm, weights 72 g and supports a 16-megapixel sensor RGB, and captures images in the green (G, 550 nm), red (R, 660 nm), red edge (RE, 735 nm), and near infrared (NIR, 790 nm) wavelengths. The sensor field of view, FOV, is 61.9 • × 48.5 • (H × V), resulting in images with a resolution of 1280 × 960 pixels of 5 cm spatial resolution. The thermal camera used was a FLIR Tau2 (Tau 2 640, FLIR Systems, Wilsonville, OR, USA), which measures 29 mm × 29 mm × 19 mm, weighs 71 g, and has a resolution of 640 × 512 pixels. This is an uncooled long wave infrared Thermal Imager covering the 7.5-13.5 µm spectral range, with a pixel size of 17 µm. The FOV (H × V) is 32 • × 26 • (H × V), resulting in images with 5 cm pixel resolution.
The spectral data retrieved from the R and NIR domains were used to compute NDVI and SAVI as indicators of vegetation development using ArcGIS (v10.2; Esri, Redlands, CA, USA). NDVI is the ratio of the difference between NIR and R and the sum of these two bands [48], while SAVI includes an additional canopy background adjustment factor, L: where NIR is reflectance in the NIR and R is reflectance in the visible red band. When the L is close to 0, the value of SAVI is equal to NDVI. However, the L factor varies inversely with the amount of vegetation present to obtain the optimal adjustment for the soil effect. In this sense, a value of L = 0.5 was considered in this study to minimize the soil noise [28].
The NDVI and SAVI algorithms take advantage of the fact that green vegetation reflects less visible light and more NIR, while sparse or less green vegetation reflects a greater portion of the visible and less NIR. These indexes combine these reflectance characteristics in a ratio making them indexes related to photosynthetic capacity and vegetative growth. Generally, NDVI and SAVI values range between −1 and +1. Only positive values correspond to vegetated zones; the higher the index, the greater the chlorophyll content of the target.
The high spatial resolution of both multispectral and thermal data allowed the separation of non-leaf material and background soil from pure leaf material. Specifically, to extract the average canopy temperature (T c ) of each measurement, the acquired thermal images were processed following the recommendations of González-Dugo et al. [35] and adjusted for an emissivity value of 0.98, which is the whole plant emissivity value reported by Jones [14]. In all cases, the average temperature of several sun-exposed leaves involved checking the visible band.
Simultaneously with the two flights, Ψ stem , ACO 2 , g s , and WUEi were determined (see Section 2.3). The maximum, minimum, and mean temperatures on the day of data collection, 19 July 2017, were 33 • C, 22 • C, and 27 • C, respectively, with no rainfall occurring in the previous weeks. Furthermore, the ET 0 , mean relative humidity (RH) and the wind speed at 2 m of the same day were 5 mm, 67%, and 4.6 km h −1 , respectively.

Vegetative Growth, Yield, and Fruit Quality
Trunk diameter was measured before harvest with a forest caliper instrument (Codimex-C 100 cm, Canada) on four trees per replicate (n = 16 trees per treatment) at a marked location about 0.3 m from the soil surface. Trunk cross-sectional area (TCSA) was estimated as being equivalent of a circle. Pruning dry mass was determined during winter dormancy in four trees per replicate (n = 16 trees per treatment). Canopy tree cover was estimated in summer with zenithal images analyzed following the procedure indicated in Conesa et al. [11].
Commercial yield at harvest (30 April, 2018) was evaluated in four trees per replicate (n = 16 trees per treatment), weighing and counting the total number of fruits per tree. Fruits affected by cracking were previously removed and were not considered in the study. Average fruit mass was calculated from total mass and number of fruits per tree. Nectarine size distribution was separated in the field by manual calibration into 7 fruit diameter categories according to [49]. Crop load was determined as the ratio of the number of fruits to TCSA. Crop water use efficiency (WUE) was determined as the ratio between yield and the total amount of irrigation applied.
Equatorial fruit diameter was assessed at harvest in 20 fruits per replicate (n = 80 fruits per treatment). Skin color was measured in the same samples using a Minolta CR-10 colorimeter (Osaka, Agronomy 2019, 9, 630 6 of 20 Japan) and the results were expressed in the CIE L*, a*, b* system, from which the skin Chroma [C* = (a*2 + b*2) 1 2 ] and Hue angle [ • hue = tan −1 (b*/a*)] were calculated. Total soluble solids content was evaluated in fruit juice, mixing 10 fruits per replicate (n = 40 fruits per treatment), using a digital refractometer (Atago ATC-1, Tokyo, Japan). Values were expressed as • Brix.

Sensitivity Analysis
Sensitivity analysis of the different plant water stress indicators was carried out using the method proposed by Goldhamer and Fereres [38], which uses the following equation: where S is the sensitivity concept, SI is the signal intensity, and CV is "noise" or the coefficient of variation. SI was calculated as the ratio between the average values of the PDI and control treatments. S is always higher than 0: the higher the value, the greater the sensitivity. Moreover, we used the corrected method (S*) described by de la Rosa et al. [39] as: In this case, an S* higher than 1 indicates sensitivity to water deficit, whereas if S* is = 0, there is no sensitivity. When S* is between 1 and 0, the results indicate that the noise is greater than the SI, meaning that there are no significant differences between the treatments studied. However, in some cases, S* might be lower than 0, indicating an anomalous behavior. As the authors [39] explained, this occurs when water stressed plants acquire values which are contrary to those expected.
All measurements were always taken in the same trees, so that variables relating to the sampling day and type and size of the sample did not interfere with the sensitivity study.

Statistical Analysis
The data were subjected to one-way analysis of variance (ANOVA) using the SPSS v 9.1 (IBM, Armonk, NY, USA) to discriminate between irrigation treatments. Statistical comparisons were considered significant at p ≤ 0.05. The degree of agreement of the regressions among variables was evaluated through the coefficient of determination (r 2 ) and the mean square error (MSE).

Environmental Conditions and Irrigation Applied
Environmental conditions were typical for Mediterranean regions characterized by hot dry summers and mild wet winters, high evaporative demand, and low rainfall. During the study period (postharvest 2017 and fruit growth 2018) rainfall was 71 mm and the ET 0 was 1068 mm. Vapor pressure deficit (VPD) reached daily mean values of −2.2 and −1.6 kPa during postharvest and fruit growth periods, respectively ( Figure 2A). Considering the ten year-seasonal average rainfall and ET 0 were 250 mm and 1320 mm, respectively [50].
The annual amounts of water applied in the growing season (2017/2018), measured by the in-line water meters were 420.4 mm and 251.6 mm for the control and PDI treatments, respectively, meaning that the soil deficit imposed in the PDI treatment represented a mean water reduction of about 40% ( Figure 2B). Environmental conditions were typical for Mediterranean regions characterized by hot dry summers and mild wet winters, high evaporative demand, and low rainfall. During the study period (postharvest 2017 and fruit growth 2018) rainfall was 71 mm and the ET0 was 1068 mm. Vapor pressure deficit (VPD) reached daily mean values of −2.2 and −1.6 kPa during postharvest and fruit growth periods, respectively ( Figure 2A). Considering the ten year-seasonal average rainfall and ET0 were 250 mm and 1320 mm, respectively [50]. The annual amounts of water applied in the growing season (2017/2018), measured by the in-line water meters were 420.4 mm and 251.6 mm for the control and PDI treatments, respectively, meaning that the soil deficit imposed in the PDI treatment represented a mean water reduction of about 40% ( Figure 2B).

Seasonal Evolution of Terrestrial Soil and Plant Water-Status Indicators
The mean θ v in the control treatment was almost constant in the top 0-0.5 m of soil, with values close to that corresponding to the field capacity (290 mm m −1 ) during the 2017/2018 growing season (data not shown). De la Rosa et al. [5] found similar θ v values (≈ 300 mm m −1 ) in full-watered early-maturing nectarine trees also cultivated in clay-loam soils. Therefore, control trees maintained REW 0-0.5 m values close to unity throughout the irrigation season, with a mean seasonal value of 0.98, while the PDI treatment showed a mean reduction in REW 0-0.5m of up to 15% during the postharvest period and 4% during fruit growth ( Figure 2C).
A non-flat pattern of the Ψ stem in both irrigation treatments reflected the climatic demand ( Figure 2A), with decreasing values from May onwards. In control trees, mean Ψ stem was −0.82 MPa ( Figure 3A), which is characteristic of well-watered nectarine trees [3,51,52], whereas in the PDI treatment Ψ stem values were on average 0.25 MPa lower with respect to control trees, this difference being statistically significant from mid-July onwards. A minimum Ψ stem value of −1.7 MPa was reached in September in the PDI treatment ( Figure 3A). Girona et al. [13] and Naor et al. [51] established threshold Ψ stem values of −1.5 MPa to ensure no impairment of bloom fertility and −2.0 MPa to limit the occurrence of double fruits for peach trees. A short-mild water deficit (Ψ stem = −1.25 MPa) before harvest might have negatively affected mid-maturing nectarine fruit size and yield [52].

Vegetative Growth, Yield, and Fruit Quality
There were no significant differences between the control and PDI treatments as regards to the yield components studied (yield, nº fruits, fruit mass, and crop load efficiency). Moreover, the crop WUE was similar for both irrigation treatments ( Table 1). The rainfall events that occurred during the fruit growth period ( Figure 2B) increased the fruit cracking, which was slightly higher in the PDI treatment (Table 1). In this respect, Galindo et al. [62] reported that rainfall intensified fruit peel cracking in water stressed pomegranate trees, the result of an The time course evolution of the gas exchange parameters was affected by the soil water deficit imposed in the PDI treatment ( Figure 3B-D). ACO 2 and g s values followed similar trends, with a Agronomy 2019, 9, 630 9 of 20 significant decrease compared with the control treatment from August to the end of the postharvest period. ACO 2 and g s patterns in the control treatment averaged seasonal values of 18 µmol m −2 s −1 and 263 mmol m −2 s −1 , respectively, while the water deficit imposed in the PDI treatment pointed to a mean reduction with respect to the control of 12% (ACO 2 ) and 18% (g s ), respectively ( Figure 3B,C). Maximum values of were noted during fruit growth period followed by a gradual decrease as the season progressed (Figure 3), which could be ascribed to the feedback effect of the fruits on leaf photosynthesis [53,54]. Some studies have indicated that a mild soil water deficit in the postharvest increases the sensitivity of leaf gas exchange parameters to environmental conditions [55]. Furthermore, the dynamics of Ψ stem ( Figure 3A) were related with the photosynthetic activity of the early-maturing nectarine trees, which could be particularly useful to identify the critical stages of the crop and the threshold depletion values for starting irrigation [56]. WUEi did not exhibit significant differences between irrigation treatments ( Figure 3D), as also found by Conesa et al. [57] in deficit irrigated table grapes, despite the fact that water shortage usually increases transpiration efficiency [58].
Similar values for these terrestrial plant water status indicators were reported by Marsal and Girona [59] and Flexas et al. [60] in peach trees, where stomatal conductance was considered as a particularly suitable plant water stress indicator. Moreover, Shackel at al. [61] found that at a Ψ stem value of about −1.5 MPa, the decrease in leaf gas exchange might be compensated by a reduction in the growth rate of the vegetative apexes, which are the major users of carbohydrates during the postharvest. In fact, in early-maturing nectarine trees, Conesa et al. [11] observed an improvement in the plant water status after removal of the water sprouts.

Vegetative Growth, Yield, and Fruit Quality
There were no significant differences between the control and PDI treatments as regards to the yield components studied (yield, n • fruits, fruit mass, and crop load efficiency). Moreover, the crop WUE was similar for both irrigation treatments ( Table 1). The rainfall events that occurred during the fruit growth period ( Figure 2B) increased the fruit cracking, which was slightly higher in the PDI treatment (Table 1). In this respect, Galindo et al. [62] reported that rainfall intensified fruit peel cracking in water stressed pomegranate trees, the result of an asymmetric increase in fruit turgor pressure because aril turgor increased to a much greater extent than peel turgor, favoring cracking. Values are means ± SE (n = 4 replicates). ns = not significant, and * = significant at p ≤ 0.05. z corresponds to commercial size B according to [49].
It should be noted that nectarine yields were lower than those reported by de la Rosa et al. [5], probably due to the differences in crop management practices (e.g., fruit thinning). Furthermore, the PDI treatment did not increase the crop WUE despite providing 40% less water than the control ( Figure 2B and Table 1), which might be explained by the low number of commercial-sized fruits per tree, due to the higher incidence of cracking in the PDI treatment (Table 1). Previous studies on deficit irrigation applied during postharvest pointed to a decrease in peach yield in the following year as a result of fewer fruits per tree, while fruit size remained relatively unchanged in an early-maturing peach [4] and nectarine [51] trees.
In our study, the soil water deficit imposed in the PDI treatment did not affect the vegetative components studied (pruning, canopy tree cover, and TCSA) ( Table 1), as also reported by de la Rosa et al. [10]. However, a reduction in tree size was noted in peach trees submitted to severe water deficits [13].
In terms of fruit quality, only skin Chroma significantly increased in the PDI treatment, indicating the lower red coloration compared with control fruits. Alcobendas et al. [63], in early-maturing peach fruits, emphasized that greater exposure to sunlight is able to compensate the possible negative effects of the fruit position on the tree when submitted to DI strategies. Figure 4 shows the mean values of the remote (NDVI, SAVI, T c , and T c -T a ) and terrestrial plant water status indicators (Ψ stem , ACO 2 , g s , and WUEi) measured by the t1 (10:00 GTM) and t2 (12:00 GTM) flights. An illustration of the indexes derived from the UAV imagery at t2 (NDVI, SAVI, T c -T a ) and the sampled trees for terrestrial indicators is shown in Figure 5. Pure vegetation NDVI values ranged from 0.76 to 0.93 and from 0.73 to 0.91, at t1 and t2, respectively, whereas lower values were obtained for the SAVI index, which varied between 0.40-0.91 (t1); and 0.36-0.64 (t2). As regards to the T c and T c -T a values, they were closely dependent on the flight time, with lower mean values at t1 (T c ≈ 23.5 • C; T c -T a ≈ −4 • C) than at t2 (T c ≈ 32.4 • C; T c -T a ≈ 0-2 • C). However, neither remote plant water status indicator identified significant differences between the control and PDI treatments, with p values of 0.25/0.26, 0.07/0.42, 0.47/0.15, and 0.47/0.14 for NDVI, SAVI, T c , and T c -T a , at t1/t2, respectively.

Remote and Terrestrial Plant-Water-Status Indicators
Canopy architecture, including leaf angle distribution and leaf area density, could also explain the variability observed in NDVI, SAVI, and T c -T a among the different treatments [64,65]. However, the methodology followed in this work to discriminate between vegetated and soil surfaces (see Section 2.4) minimized the effect that open-center canopies could have been influenced these variables, since the non-vegetated areas inside the tree crown were removed from the analysis.
Ψ stem at both t1 and t2 flight times was the only plant water status indicator (considering both remote and terrestrial) that detected significant differences between irrigation treatments, with mean Ψ stem values of −0.74 and −1.01 MPa, for control and PDI treatments, respectively, at t1, and −0.88 and −1.34 MPa for control and PDI treatments, at t2. In this respect, it is important to note that the use of Ψ stem for scheduling deficit irrigation strategies has been widely used in many deciduous crops such as nectarine [5] and peach [66]. These authors reported the feasibility of using Ψ stem as the plant water status indicator, not only for its robustness but also for its stability in successive growing seasons. Moreover, a model based on soil water content and meteorological variables that provides information on plant water status has been proposed as a guide for irrigation scheduling of early-maturing peach trees under Mediterranean conditions [67]. Agronomy 2019, 9,  Tc-Ta (ºC)

Role of NDVI, SAVI, and Tc-Ta indexes
One direct use of NDVI and SAVI is to characterize canopy growth and tree vigor [48,71,72], and so these indexes can be good estimators of pruning needs. However, it is well known that the lack of correlation between NDVI and vegetative growth could be due to soil water distribution and soil surface anisotropy, as well as the angular geometry of illumination at the time of the measurements [73]. Rondeaux et al. [74] reported that NDVI was sensitive to soil background noise and concluded that it is difficult to interpret when the vegetation cover is low. For this reason, a relationship between pruning and canopy tree cover ( Figure 6A) and NDVI and SAVI ( Figure 6B,C) was drawn with data referring to the entire planting framework (6.5 m × 3.5 m) rather than considering only pure vegetation.
The results indicated that canopy tree cover and the vegetation indexes (NDVI and SAVI) were related to changes observed in the pruning removed in the control and PDI treatments. When data were pooled, all of them showed a linear relationship with pruning weight, with a coefficient of determination (r 2 ) of 0.42, 0.63, and 0.51 for canopy tree cover, NDVI, and SAVI, respectively. Dobrowski et al. [75] found in vineyards that NDVI was linearly correlated with field-wide measurements of pruning weight density. Hogrefe et al. [76] found a strong curvilinear relationship between NDVI and the biomass-the more biomass observed, the higher the NDVI value. Similar to the remote plant water status indicators, the leaf gas exchange indicators (ACO 2 , g s , and WUEi), did not reflect significant differences between the two irrigation treatments ( Figure 4F-H). This fact could be due to the anisohydric behavior of nectarine trees, which tolerate soil drought and responds to a decrease in water availability by tissue dehydration [68]. Indeed, anisohydric plants have more variable leaf water potentials and maintain their stomata open longer periods, accompanied by high photosynthetic rates, even in the presence of decreasing leaf water potentials or increasing atmospheric water demands [69]. This also agrees with the g s versus T c -T a relationship found in Citrus, where T c -T a did not vary when g s was greater than a mean value of 200 mmol m −2 s −1 [70]. Therefore, the moderate plant water stress induced by the PDI treatment caused a Ψ stem reduction (0.27 MPa at t1 and 0.46 MPa at t2) while maintaining g s values almost constant. This would also explain why, according to the average g s range observed in the PDI treatment (210 and 180 mmol m −2 s −1 at t1 and t2, respectively), no differences were observed between treatments when the T c -T a indicator was used ( Figure 4D,G).

Role of NDVI, SAVI, and Tc-Ta indexes
One direct use of NDVI and SAVI is to characterize canopy growth and tree vigor [48,71,72], and so these indexes can be good estimators of pruning needs. However, it is well known that the lack of correlation between NDVI and vegetative growth could be due to soil water distribution and soil surface anisotropy, as well as the angular geometry of illumination at the time of the measurements [73]. Rondeaux et al. [74] reported that NDVI was sensitive to soil background noise and concluded that it is difficult to interpret when the vegetation cover is low. For this reason, a relationship between pruning and canopy tree cover ( Figure 6A) and NDVI and SAVI ( Figure 6B,C) was drawn with data referring to the entire planting framework (6.5 m × 3.5 m) rather than considering only pure vegetation. Agronomy 2019, 9, x FOR PEER REVIEW 14 of 22 Canopy tree cover (%) 40 50 60 70 Pruning (kg tree One indirect use of NDVI and SAVI is to detect plant water stress situations. Indeed, both remote indexes have shown positive correlations with Ψstem and gs in several crops [33]. Nevertheless, as can be seen in Figure 4A,B, neither NDVI nor SAVI identified significant differences between the control and PDI treatments. When both remote indexes were related with Ψstem, the linear relationship showed a poor coefficient of determination (r 2 = 0.27, p ≤ 0.05, and r 2 = 0.22 ns) ( Figure 7A,C).
The lack of correlation between NDVI and SAVI with Ψstem at t1 and t2 flight times might be related to the moderate soil water deficit induced by PDI treatment based on real-time SWC. Although the PDI treatment was able to promote a noticeable water saving (about 40%) with respect to the control treatment without affecting the yield components studied (Table 1), the SWC thresholds imposed were not sufficient to promote significant differences between the treatments in the structural remote indexes (NDVI and SAVI). Thus, in order to ascertain water stress conditions using NDVI and SAVI it would be necessary to apply more severe and longer periods of water stress.
The relationship between Ψstem and Tc-Ta explained the 56% of the variations observed ( Figure 7E), which suggests that Tc-Ta describes the response to water stress better than the leaf structural changes assessed by NDVI and SAVI indexes. Therefore, the spectral VIS-NIR images were not as sensitive to water deficits as those derived from canopy temperature [36,77].
In our study, the Tc-Ta behavior coincided with that observed for gs both at t1 and t2 ( Figure 4D,G) since the major determinant of leaf temperature is the rate of evaporation or transpiration from the leaf [36]; thus, as plants transpire, the temperature of the leaves did not increase as usually happens under drought conditions [21]. However, it is also known that the variability of Tc in deciduous trees is different from that observed in annual crops, which mainly results from the heterogeneity of soil properties and the lack of irrigation uniformity when using UAV thermal cameras [78]. Also, the heterogeneity of the Tc in moderate water stress conditions was explained by González-Dugo et al. [65,79] as being due to the onset of stress in a few areas within the crown with substantial stomatal closure, while in the rest of the crown the stomata could still be open.
Camino et al. [80] pointed to the importance of using high resolution hyperspectral and thermal imagery for pure-object segmentation extraction from tree crowns in order to ascertain water stress situations. In this sense, Bellvert et al. [18] recommended an optimum pixel size to detect water stress in peach and nectarine orchards that ranged from 0.6 to 0.8 m in images taken from UAV. The results indicated that canopy tree cover and the vegetation indexes (NDVI and SAVI) were related to changes observed in the pruning removed in the control and PDI treatments. When data were pooled, all of them showed a linear relationship with pruning weight, with a coefficient of determination (r 2 ) of 0.42, 0.63, and 0.51 for canopy tree cover, NDVI, and SAVI, respectively. Dobrowski et al. [75] found in vineyards that NDVI was linearly correlated with field-wide measurements of pruning weight density. Hogrefe et al. [76] found a strong curvilinear relationship between NDVI and the biomass-the more biomass observed, the higher the NDVI value.
One indirect use of NDVI and SAVI is to detect plant water stress situations. Indeed, both remote indexes have shown positive correlations with Ψ stem and g s in several crops [33]. Nevertheless, as can be seen in Figure 4A,B, neither NDVI nor SAVI identified significant differences between the control and PDI treatments. When both remote indexes were related with Ψ stem , the linear relationship showed a poor coefficient of determination (r 2 = 0.27, p ≤ 0.05, and r 2 = 0.22 ns) ( Figure 7A,C).
The lack of correlation between NDVI and SAVI with Ψ stem at t1 and t2 flight times might be related to the moderate soil water deficit induced by PDI treatment based on real-time SWC. Although the PDI treatment was able to promote a noticeable water saving (about 40%) with respect to the control treatment without affecting the yield components studied (Table 1), the SWC thresholds imposed were not sufficient to promote significant differences between the treatments in the structural remote indexes (NDVI and SAVI). Thus, in order to ascertain water stress conditions using NDVI and SAVI it would be necessary to apply more severe and longer periods of water stress.
The relationship between Ψ stem and T c -T a explained the 56% of the variations observed ( Figure 7E), which suggests that T c -T a describes the response to water stress better than the leaf structural changes assessed by NDVI and SAVI indexes. Therefore, the spectral VIS-NIR images were not as sensitive to water deficits as those derived from canopy temperature [36,77].
In our study, the T c -T a behavior coincided with that observed for g s both at t1 and t2 ( Figure 4D,G) since the major determinant of leaf temperature is the rate of evaporation or transpiration from the leaf [36]; thus, as plants transpire, the temperature of the leaves did not increase as usually happens under drought conditions [21]. However, it is also known that the variability of T c in deciduous trees is different from that observed in annual crops, which mainly results from the heterogeneity of soil properties and the lack of irrigation uniformity when using UAV thermal cameras [78]. Also, the heterogeneity of the T c in moderate water stress conditions was explained by González-Dugo et al. [65,79] as being due to the onset of stress in a few areas within the crown with substantial stomatal closure, while in the rest of the crown the stomata could still be open.   Camino et al. [80] pointed to the importance of using high resolution hyperspectral and thermal imagery for pure-object segmentation extraction from tree crowns in order to ascertain water stress situations. In this sense, Bellvert et al. [18] recommended an optimum pixel size to detect water stress in peach and nectarine orchards that ranged from 0.6 to 0.8 m in images taken from UAV.
Interestingly, when the mean VPD value registered at t1 (−0.98 kPa) and t2 (−2.96 kPa) was included in the independent term, along with the Ψ stem, as a two-variable function of the relationship with NDVI ( Figure 7B), SAVI ( Figure 7D), and T c -T a (Figure 7F), the coefficient of determination (r 2 ) improved considerably in this order: T c -T a > SAVI > NDVI. As VPD integrates T a and RH, it is a sensitive agro-meteorological variable for correlating with Ψ stem [43]. In peach trees, Abrisqueta et al. [81] proposed an alternative to the field measurement of Ψ stem , using a multiple linear regression equation based on SWC, mean VPD, and growing degree hours (GDH) values. These authors reported that the contribution of the soil and atmosphere components to Ψ stem differed according to the intensity of the water deficit imposed in each irrigation treatment. Blanco et al. [82] also proposed a multiple linear regression equation based on average soil water tension and mean VPD for estimating Ψ stem in sweet cherry trees.

Sensitivity of Remote and Terrestrial Plant Water Status Indicators
The sensitivity analysis at t1 and t2 flight times showed that Ψ stem was the plant water status indicator with the highest signal intensity (SI) followed by T c and T c -T a ( Table 2). Indeed, T c -Ta registered the highest SI values of the remote sensing indicators studied. Although NDVI had lower signal intensity than Ψ stem , its sensitivity values (S) were much higher due to the low CV (0.99% and 1.10% at t1 and t2, respectively). Blanco et al. [82], in sweet cherry trees, reported the high sensitivity of the soil matric potential (Ψ m ) despite the high CV. In our case, the variability observed in NDVI, SAVI, and T c -T a values may have been mediated by the environmental conditions, underlining the importance of considering VPD along with the plant water status ( Figure 7B,D,F). For this reason, the S* method [39], which diminished the influence of CV in the analysis, showed the strongest sensitivity for Ψ stem , followed by T c -T a at both flight times ( Table 2). Table 2. Sensitivity analysis for terrestrial and remote plant water status indicators at the two flight times: t1 (10:00 GMT) and t2 (12:00 GMT). Values are means of four replicates. SI, signal intensity; CV, coefficient of variation; S, SI/CV; S*, (SI-1)/CV; Ψ stem , midday stem water potential (MPa); ACO 2 , net CO 2 assimilation rate (µmol m −2 s −1 ); g s , stomatal conductance (mmol m −2 s −1 ); WUEi, instantaneous water use efficiency (µmol mmol −1 ); NDVI, normalized difference vegetation index; SAVI, soil adjusted vegetation index (SAVI); T c , canopy temperature ( • C); T c -T a , canopy to air temperature difference ( • C).

Conclusions
The current work showed that in early-maturing nectarine trees submitted to moderate deficit irrigation, Ψ stem is a valuable robust indicator for detecting water stress; however, it needs to be considered together with VPD for comparison with remote sensing data to reveal spatial patterns of water stress. The described relationship was able to predict remote sensing indicators when they were not available at the time of Ψ stem acquisition. Therefore, Ψ stem is still a powerful indicator of the water status of the plant, which must be taken into account to adjust the adequate soil water depletion at each phenological period to apply the precise water needs. Furthermore, Ψ stem followed by T c -T a registered the highest signal intensity for detecting water deficit situations. NDVI and SAVI assessed in the postharvest period were seen to be good indicators for estimating the winter pruning needs. However, they were not able to identify significant differences between control and PDI treatments, since these multispectral indexes respond less strongly to water stress than T c -T a . These results suggest that care should be taken when NDVI and SAVI are used to assess moderate water deficits in early-maturing nectarine trees. More severe and/or longer water stress conditions of stress are probably needed. Whatever the case, precise deficit irrigation based on SWC used 40% less total irrigation volume than a traditionally scheduled treatment with no penalty in yield of early-maturing nectarine trees. This suggests that the use of real-time threshold SWC values could be a promising irrigation strategy in clay-loam soils in Mediterranean areas endangered by climate change.